摘要
深度学习的日益流行和不断的突破,积极地推动了估算锂离子电池荷电状态(SOC)新方法的研发。常用于文本翻译、语音识别等领域的循环神经网络算法已开始应用在电池SOC估算领域。概述了几种常用于电池SOC估算的循环神经网络算法,并根据结构复杂度将其分为简单、深度以及扩展循环神经网络算法。此外,从模型输入变量、数据预处理、循环神经层数及其神经元数、优化器、损失函数、测试文件、研究的温度、使用的电池类型以及估算误差方面对这些循环神经网络算法进行了比较,总结了这几种估算方法的优缺点,让读者对循环神经网络算法有更直观的了解。循环神经网络算法与其他算法相结合来估算电池SOC,将会成为重点研究方向。
The increasing popularity and continuous breakthroughs of deep learning have actively promoted the development of new methods for estimating the state of charge(SOC)of lithium-ion batteries.The recurrent neural network algorithm commonly used in text translation,speech recognition and other fields has begun to be applied in the field of battery SOC estimation.This article outlined several recurrent neural network algorithms commonly used in battery SOC estimation,and divided them into simple,deep,and extended recurrent neural network algorithms based on structural complexity.In addition,these recurrent neural network algorithms were implemented in terms of model input variables,data preprocessing,number of circulating neural layers and their neurons,optimizer,loss function,test file,temperature studied,battery type used,and estimation error.In comparison,the advantages and disadvantages of these estimation methods were summarized,so that the reader has a more intuitive understanding about the recurrent neural network algorithm.The combination of recurrent neural network algorithm and other algorithms to estimate battery SOC will become a key research direction.
作者
徐帅
刘雨辰
周飞
XU Shuai;LIU Yu-chen;ZHOU Fei(National Key Laboratory of Helicopter Transmission Technology,Nanjing University of Aeronautics and Astronautics,Nanjing Jiangsu 210016,China)
出处
《电源技术》
CAS
北大核心
2021年第2期263-269,共7页
Chinese Journal of Power Sources